Core concepts
A handful of ideas explain almost everything Acteme does: how a project is organized, what a line's status means, how takes are chosen, when a render is free and how credits are charged.
Projects, scenes and lines
A project is one game or one voice-over pass. It holds the cast, the script, the languages you ship and the output spec every file is rendered to.
- Characters belong to a project. Each has a voice, an optional default style and optional per-language voices.
- Scenes group lines. They're optional, show up as collapsible groups in the script, and can be rendered on their own. The default file template puts each scene in its own folder.
- Lines are the unit of work. A line has a key, a character, the text to speak, an optional style note, writer notes and one translation per extra language.
- The lexicon holds pronunciation respellings for the whole project.
Keys
A line's key is its permanent identity: re-imports match on it, file names are built from it, and your game looks clips up by it. Keys may contain letters, digits and . _ : - /, up to 120 characters. Treat them like asset GUIDs and never reuse a key for a different line.
Line status
Every line has a status, tracked separately for each language.
| Status | Meaning |
|---|---|
draft | Nothing rendered yet. |
rendered | Has at least one take that matches the current text, style and voice. |
approved | You signed off on it. Approved lines are skipped by "Stale and draft" renders. |
stale | Something that changes the audio changed after the last render. |
A rendered or approved line becomes stale when:
- its text, style or character changes (by hand, by bulk edit or by re-import);
- its character's voice, default style or model changes, for the languages where the audio would differ;
- a lexicon entry that affects its text is added, edited or removed;
- for a translation: the translated text or its style changes.
Draft lines stay draft, since they have nothing to go stale.
Takes
Each render of a line in one language produces a take. Takes are numbered per line and language, and a line can collect up to 10 per render request.
Exactly one take per line and language is selected. That's the take that ships: it's what the ZIP export, the manifest and the engine plugins deliver. The first take a line gets is selected automatically. Later takes are added alongside it, and the selection only changes when you pick another take.
Caching
If you've already rendered identical audio (the same line, voice, direction and settings), it's reused instead of rendered again. The item is reported as cached and costs nothing.
Each take number is a separate performance, so asking a line for another take always renders new audio. Cache hits come from identical work done elsewhere: the same bark used by several lines, a line that was deleted and imported again, or a repeated one-off request in the playground.
Credits
Renders, machine translation and one-off speech cost credits. Roughly, one credit is one spoken character on the Pro model; Credits & plans has the exact formula.
- Reserve. When you queue a render, the full cost is reserved from your balance up front. If your balance is too low, nothing is queued.
- Render. Each line is rendered independently and retried automatically on temporary errors.
- Settle. Lines that render successfully keep their charge. Lines that come back cached, fail or are canceled are refunded automatically.
Every charge and refund is written to the ledger under Plans & billing.
Jobs
A render request becomes a job with one item per take to render. Jobs run in the background; you can close the tab. Their status is one of queued, running, completed, partial (some items failed), failed or canceled. How many jobs run at the same time depends on your plan. See Rendering & jobs.